- Home
- Skills
- Business & Operations
- Draft Inbox Replies a Human Approves
More screenshots
Works with the AI Agents you already use
Draft Inbox Replies a Human Approves
Your agent drafts replies to the questions your inbox gets every week — from answers you approved word for word — and never sends anything.
$12
Draft Inbox Replies a Human Approves
Example session with this skill installed
Check the new inbox message from a customer asking about our refund policy against the approved FAQ set and draft a reply if you are confident, otherwise route it for manual review.
- Read your context and instructions
- Compiled the draft inbox replies
Result
Match: Refund policy FAQ — confidence 0.92 (threshold 0.80)
Decision: Draft only — nothing was sent.
Draft reply: "Thanks for asking. Refunds are available within 30 days of purchase for unused items, and we credit the original payment method within 5 business days. You can start a return from your order page." Ready for your review and edit before sending.
Idempotency: Message ID msg_4821 already recorded as processed, so re-running this inbox scan will not draft a second reply.
Audit trail: Decision, confidence, FAQ ID, and message ID logged. No message body was copied into the audit log.
If confidence had been below threshold: the message would have been routed to manual review with the closest FAQ candidates listed, instead of drafting.
Connects securely to your tools. The creator never sees your data.
About this skill
Most inbox automation fails in one direction: it answers something it should not have, confidently, and sends it. This skill is built around the opposite failure mode. It matches incoming mail against a small set of answers you approved in advance. A strong, clear match becomes a draft in your inbox. Everything else goes to a person, untouched. If you remember one rule, make it this one: this system creates drafts; a human sends.
Inside the package: the skill file covering the three-stage design, the threshold rules, and the never-answer categories; a filled example FAQ set; and a reference on the audit trail and duplicate protection. The design keeps three jobs in three boxes with hard walls between them — classify the message, draft from the approved answer, deliver. Delivery is not this system's job: there is no send step anywhere in its path, and the mail access it needs is read-and-draft only, never send permission "just in case."
A run goes like this: you approve the FAQ set once — each entry is a question pattern plus the exact answer text, word for word, and later edits to an answer go through the same approval as the original. The agent scores each incoming message against every entry. A draft is created only when the best match clears the threshold you set and clearly beats the runner-up; a close score between two answers is treated as information — a person should read this — not a coin flip. Complaints, legal questions, pricing disputes, health questions, and anything personal route straight to manual review whatever they resemble. Each message is claimed in a durable store before processing, so a restart or a second run can never produce a duplicate draft. And the audit trail records decisions and message IDs without copying message bodies into the log. You review the drafts and send the ones you stand behind.
Find more skills like this on the Claude Skills Marketplace, the Codex Skills Marketplace and the Gemini Skills Marketplace, or browse Claude Skills for Business & Operations.
How to use
- 1
Connect your agent to the Agensi MCP
Copy the prompt and paste it into Claude, ChatGPT, Cursor or any agent that supports MCP. It sets up the connection for you. Manual setup
- 2
Sign in to Agensi
Your agent shows a sign-in link. Sign in and click Allow.
- 3
Ask for what you need
Your agent loads this skill and uses it on your task. It always asks before buying a skill.
Reviews
No reviews yet
Be one of the first to try it. Every listed skill goes through the trust checks below.
Security scanned
Checked by our 8-point scan before listing
Fresh listing
Recently published to Agensi
Money-back guarantee
Doesn't do what its page promises? You get your money back.
Trust & safety
Security scan
Security scan passed 6 days ago
- Software
- An approved FAQ set and inbox access